Skip to content

Methodology

How a ballot becomes a number

CitizenVote is a public record, not a campaign and not a probability poll. This page is the method. If a chart and this page disagree, this page wins.

What this is

The Constitution opens with the people. CitizenVote takes that seriously and narrowly: it asks specific questions, records one vote per account, and shows the count. It does not tell you what the country “really” wants beyond the people who voted here.

The site is nonpartisan in the only way that can be checked. Questions are propositions, not slogans. Each one carries a short explainer and two columns of arguments, given the same space. Result charts use teal for support and copper for opposition so the picture does not borrow party colors. CitizenVote is not affiliated with any party, campaign, or government. The public address is CitizenVoteUSA.com.

How a question is written

A useful question can be supported or opposed. “Abortion” is a subject. “Keep abortion legal nationwide in the first trimester” is a proposition. The explainer states the current rule in plain language and says what the proposition does not decide.

Arguments are claims people actually make, shortened, and not graded. A column is not an endorsement. Administrators can add and edit questions. Today a ballot is attached to the question as it reads now. If the wording changes, older ballots are not relabeled. Edit with care. A later version should freeze the wording beside each ballot.

One person, one vote

Each account has one row per question. Changing your mind updates that row. It does not create a second ballot. The same rule applies to debates: one upvote per account per matchup, and you may withdraw it.

Casual duplication is made more expensive, not impossible. Account creation, sign-in, voting, verification codes, and debate proposals are rate-limited by account and by network address. The signup form includes a hidden field that automated clients often fill. These controls do not stop a determined person with many email addresses. That is what the verification ladder is for: readers can raise the bar when they read a result.

Verification levels

Every result can be filtered to accounts at or above a level.

  1. Unverified. An account exists. The email has not been confirmed.
  2. Email verified. The person confirmed a code sent to the address on the account.
  3. Phone verified. The person confirmed a code sent to a U.S. phone number.
  4. Identity verified. A document check has been recorded. No identity vendor is connected. This level is a placeholder.

The checks sit behind one interface, VerificationProvider. When an email provider is configured, the code is sent to the address on the account and is not shown on screen. A local copy with no email key uses a stub instead: the code appears on the account page, labeled as a preview. Phone and identity checks are not connected to a vendor. An administrator can record an identity check for a demonstration. That recording is not a document review, and the admin screen says so.

Raw results

A raw result is the share of ballots in the current filter. “Support” adds “strongly support” and “somewhat support.” “Oppose” adds the two oppose choices. “Neither” and “Not sure” are reported on their own and stay in the denominator. The margin is support minus oppose, in percentage points. Leaving the middle choices in the base keeps a large undecided group from disappearing into a false landslide.

Grouped results — age, sex, party, Census region, and state — use the same definitions. A group with fewer than five ballots is shown as a count only. Five people are not a state consensus, and the page will not pretend they are. Party is self-described. It is a breakdown, not a fact from the Census.

Weighted results

People who create an account are not a miniature of the country. Online samples often skew younger. Weighting is a correction on three margins the Census publishes for adults: age, sex, and region. It is called raking, or iterative proportional fitting.

Suppose older adults are about 23 percent of the adult population and 11 percent of the ballots. Each older ballot is scaled up, and younger ballots are scaled down, until the weighted sample matches the published age shares — and, in turn, the sex and region shares. The adjustment rotates through the three margins, because fixing one nudges the others. After 30 passes, weights are limited so one person counts as no more than five ballots and no less than one-fifth, then rescaled so the average weight is 1.

Only ballots with age, sex (female or male), and a state are eligible. “Prefer not to say,” and any blank, stays in the raw count and drops out of the weighted count. If a whole population group is missing, that group is dropped from the target and the rest of the margin is rescaled. The results page says when that happens. The effective sample size, a standard summary of uneven weights, is shown beside the weighted figure. It is lower than the raw count when some ballots have been scaled far from 1.

Party is not a target. There is no census of party identification to rake toward, and inventing one would pretend to a precision the record does not have.

Rounded shares of the U.S. population age 18 and older, based on Census Bureau population estimates. Sex is the two-category adult split. Region is the four Census regions (the District of Columbia is in the South). These are published targets for weighting, not a live feed from a Census API.

Age
18–2411.7%
25–3417.4%
35–4416.7%
45–5415.5%
55–6416.1%
65 or older22.6%
Sex
Female51.1%
Male48.9%
Census region
Northeast17.2%
Midwest20.6%
South38.6%
West23.6%

What weighting does not do

It does not make this a scientific poll. It does not fix a person who misstates their age or their view. It does not create respondents who never heard of CitizenVote. Read a weighted number as: among people who voted here and reported age, sex, and state, this is the result after those three margins are aligned to the published adult population. Read a raw number as the straightforward count. The page shows both.

Demonstration ballots

The first database includes sample accounts so a reviewer can see charts, filters, and weighting before any public launch. Their names begin with “Sample voter,” their email addresses are on demo.citizenvoteusa.com, and they cannot sign in. Their ballots are illustrative. The shape is varied by age, sex, and self-described party so the instrument has something to show. It is not a copy of any published poll, and it is not the public.

Every result that includes them says so. “Live accounts only” removes them. Do not present a chart that still includes them as the country’s opinion.

Debate rankings

A debate proposal is two public figures, an optional topic, and an optional link to a question already on the record. Other accounts upvote. The board ranks matchups by those upvotes. The same verification filter used on issues applies here. One account, one upvote, withdrawable.

Debate rankings are not weighted. An upvote is an expression of interest, not a demographic sample, and weighting it would imply a representativeness it does not have. Example matchups were written to span the spectrum. The order of names is not an endorsement, and listing a person is not a claim about them.

Privacy

CitizenVote stores the account, the password as a hash, the ballot, any verification stamps, and the optional profile: state, age bracket, sex, and party. Public pages show aggregates. They do not list a name beside a vote. Small groups are suppressed. Profile fields can be left blank. Selling personal information is not part of this record.

A note on the address

The canonical site is https://citizenvoteusa.com. Pages declare that address so a shared link points at CitizenVoteUSA.com even when you are reading a local copy.

Methodology · CitizenVote